A Hybrid Explainable Deep Learning Framework for Reliable and Human-centric AI

Deep learning systems impact critical decisions yet remain opaque. This paper presents a hybrid XAI framework combining global rule extraction with local gradient-based attribution. We introduce the Human-Centric Reliability Index (HCRI) validated across trust, satisfaction, and actionability. Evaluations over medical, credit, and navigation domains achieve 96.2% accuracy and 0.92 HCRI, outperforming LIME, SHAP, and attention baselines. A structured XAI taxonomy, dataset limitations, algorithm benchmarks, multimodal challenges, and deployment constraints are also presented.